"Low-magnification" telecentric

[aphi]

The idea is essentially to have the camera on a gantry do focus stacks and then stitch the focus stacks together in order to create high-resolution "scans" of very non-flat objects.

For that to work the angle of view needs to be ~0, so telecentric. Otherwise each stack has its own vanishing point.

I'm looking for something in the 1:1 to 1:3x range (anything lower magnification would become impractically large) on a full-frame sensor.

The only lens that I know of that might fit the bill is the Sigma 180mm f2.8 APO EX DG OS HSM. According to their optical scheme in a patent drawing this lens ought to be telecentric at around 1:1.3: https://www.photonstophotos.net/General ... sO,OffAxis

But it's pretty large (2kg) and relatively high magnification and has substantial lateral CA. Anyone has any other ideas? (Also perhaps to the general approach outlined above, maybe this can be done without a telecentric lens after all?)

[Lou Jost]

That's interesting information about the Sigma lens!

An alternative is to put a close-up lens in front of any macro lens, adjusting the distance between the close-up lens and the macro lens until the combo is telecentric. Rik has written a lot about this on the forum.

See my tests of this here:
http://www.photomacrography.net/forum/v ... highlight=
Look at the Oct 22 posts.

[rjlittlefield]

Also perhaps to the general approach outlined above, maybe this can be done without a telecentric lens after all?

To get clean stitches of arbitrary 3D objects, you definitely have to hold the entrance pupil constant for all tiles. One way to do that is to rotate around an entrance pupil that is located at some finite distance. Another way is to push the entrance pupil to infinity so that simple shift is equivalent to rotating around the pupil. A third way is to fix the lens in place and move a sensor in the image field to do both the focus stacking and the side-by-side tiling.

All of these methods are difficult to pull off well enough that in the end you get all the quality you had hoped for.

Rotating around a finite entrance pupil works well from the standpoint of image processing, but the mechanics can be fiddly. If you want to pursue that path, then I suggest to get a good pano head and carefully set it up so that the entrance pupil is at the exact center of rotation. That detail might be simplified by running the lens wide open and stopping it down with an added aperture placed as close as possible to the front element, so that the added aperture becomes the entrance pupil. Alternatively, if you're clever with robotics then you can combine shift and rotate of the camera+lens so as to hold the entrance pupil constant even with optics that place the entrance pupil awkwardly far away from the camera+lens mount.

Pushing the entrance pupil to infinity is another way of saying "telecentric optics", and there the wrinkle is that for best results you need to have the optics be telecentric over the entire field. In order to accomplish that, the front element of the lens system has to be at least as large as the subject field diameter plus the diameter of the entrance cone at the place where it enters that first element. If the front element is not that large, then the center of the field will be telecentric, but due to clipping of off-center cones, off-center parts of the field will become progressively non-telecentric. They will also go dark from vignetting, and they can develop weird aberrations due to lens elements being optimized for other light paths. See viewtopic.php?t=39724 ("Telecentric optics on the edge") for some discussion of these issues. Limitation of the telecentric field can be seen even with the Sigma lens that you link. If you focus the lens to 1:1.26 where P=11549.51, leave the lens wide open, and look closely at the green ray fan, you'll see that on the front of the lens the central ray is not parallel to the optical axis, and at the physical aperture the fan does not fill the aperture. Those are symptomatic of the same problem: the front lens element is not large enough to be telecentric in the corners at that aperture setting. To make it be telecentric across the entire field, you would have to stop down to around f/12 (nominal) so you'd be mildly into diffraction territory even if the optics were otherwise perfect. Sometimes you can do better by assembling your own telecentric combo, for example as shown at viewtopic.php?t=18323 "Telecentric combo at 0.8X to 1.69X", which is one of the posts that I'm sure Lou had in mind. See also viewtopic.php?t=1472 for some general theory.

The option to use a fixed lens and move the sensor around in the image field is one that I've never played with but Lou has. I'll defer to him for pros and cons.

I feel compelled to ask one question: are you committed to stitching, or might you simply use a high-megapixel sensor and a lens sufficiently good to shoot your subject in one stack? I realize that this approach is a lot less exciting, but if you're interested in the images and not the process, it's also by far the lowest frustration route to success.

One final issue... To continue this discussion, please clarify what size subjects and what magnification range you're thinking about. Where you write "1:3x", I'm not sure if you're talking optical magnification of 3X or 0.333X. Much of what I've written has been motivated by the 0.333X end of things.

--Rik

[rjlittlefield]

One more reference... If you mean 1X to 3X then your description sounds a lot like GIGAmacro's rig, and in that case I suggest to look at https://gigamacro.com/toolmakers-lenses/ . I'm pretty sure that the Nikon Toolmakers' objectives they use are not strictly speaking telecentric across the whole field on full frame at 1X, but apparently they're close enough to get by.

--Rik

[Lou Jost]

...The option to use a fixed lens and move the sensor around in the image field is one that I've never played with but Lou has. I'll defer to him for pros and cons....

--Rik

For my usage the option to use a fixed lens and move the sensor is very attractive because it produces nearly perfect tiles. But it is limited in resolution because the lens aperture is usually f/5.6 or smaller. I would use it for subjects 15cm across or larger.

More resolution can be obtained by using the telecentric microscope objectives mentioned by Rik, at the cost of much more work. Those Nikon MM or toolmaker's objectives come in various magnifications, with the 10x and higher versions being very good across FF and with good NA for the magnification. They do tend to make purple fringes, which often stack out. The extension distance has to be exactly right for them to work as designed. To remove any persistent purple fringes, you may want to try to filter the light source with an approx. 450nm high-pass filter.

[soldevilla]

Hello.

I have some experience in this. I have worked for almost three years, intermittently, to create a gigapixel image of the Nebra Disk from partial images. The goal was to achieve a resolution of approximately 0.01mm per pixel, which means a final file of 35,000 x 35,000 pixels.

None of the software I tested for producing the final image as a panorama worked for me.

A new approach was to use a photogrammetry program. Using such software made it easy to obtain an orthophoto, which is exactly the result I was looking for.

[rjlittlefield]

A new approach was to use a photogrammetry program. Using such software made it easy to obtain an orthophoto, which is exactly the result I was looking for.

Which photogrammetry program did you use?

--Rik

[soldevilla]

A new approach was to use a photogrammetry program. Using such software made it easy to obtain an orthophoto, which is exactly the result I was looking for.

Which photogrammetry program did you use?

--Rik

Agisoft metashape